Research On Abnormal Trajectory Detection Of Taxi Based On Data Mining Technology

With the development of intelligent transportation system and data mining technology,taxi trajectory anomaly detection based on data mining technology is an important method for realizing modern traffic intelligent monitoring,so as to automatically detect some abnormal behaviors in the process of vehicle travel and alert the drivers.Currently,there are two different levels of space-time data anomalies detection: 1)abnormal taxi trajectory;2)abnormal traffic conditions.This paper conducts an in-depth study on the abnormal behavior of taxis.The paper will carry out anomaly detection methods from these two levels.However,due to the characteristics of large amount of data,inconspicuous features,and susceptibility to interference,data mining methods have some challenged.Based on the above analysis,this paper makes full use of the different characteristics of the trajectory,and proposes a practical anomaly detection method for three different levels of anomalies.The main contents include:First,the vehicle track processing and map matching.The paper analyzes the collected data,selects and filters the GPS trajectory from the preprocessing data.From the GPS raw data,we extracts effective latitude,longitude,time,and heading information,and integrates the road network information.We use multi-weight trajectories and road network matching methods to match the GPS trajectory points to roads.Second,vehicle abnormal driving behavior detection based on GPS data mining.According to the existing taxi GPS data,the global characteristics such as speed,acceleration,direction,rotation angle and flameout time are extracted and the corresponding 13 statistics are extracted to construct the characteristic attributes of the vehicle driving behavior.We analyze and compare common similarity measures,and propose similarity measures based on structural distances and construct feature similarity matrix.In the aggregation-based hierarchical clustering of features,the Laplace mapping idea of spectral clustering is introduced to reduce the dimension of clustering data and obtains clusters' number.We propose trajectory point clustering anomaly detection method based on multi-features fusion.Finally,the clusters are divided into normal or abnormal clusters according to the size.The clusters are compared with the measured data to detect the abnormalities.Third,an abnormal detection of traffic conditions based on taxi GPS data mining.A large number of taxi GPS trajectories reflect the spatial and temporal patterns of the entire traffic situation,and the behaviors of taxi abnormal trajectories or abnormal driving paths are detected and analyzed.This study analyzes the characterization parameters of traffic flow conditions,and selects the characteristics to represent the abnormal traffic condition.In this procedure,we establishes a traffic abnormality detection model and a prediction model,proposing a regional road speed model and the improved fuzzy C-means simulated annealing genetic algorithm data stream clustering model.This study combines the spatial-temporal characteristics of GPS trajectory data with driver behavior,establishes evaluation criteria for abnormal characteristics,proposes an abnormal behavior evaluation index CE,and compares the set trajectory anomaly thresholds to determine the abnormality of path selection abnormal behavior.Finally,traffic anomalies were analyzed from the anomaly detection results.Finally,taking the taxi GPS data of Xi'an as the data object to verify the anomaly detection method is proposed in this paper.The proposed method is compared with the classical algorithms from the aspects of detection accuracy,recall rate,and so on.The experimental results show that the method proposed in this paper can detect taxi abnormal behaviors and abnormal traffic behavior efficiently and accurately.